Supplementary Information for ‘A pre-Campanian Ignimbrite techno-cultural shift in the Aurignacian sequence of Grotta di Castelcivita, southern Italy’
Armando Falcucci, Simona Arrighi, Vincenzo Spagnolo, Matteo Rossini, Owen Alexander Higgins, Brunella Muttillo, Ivan Martini, Jacopo Crezzini, Francesco Boschin, Annamaria Ronchitelli, Adriana Moroni
armando.falcucci@uni-tuebingen.de (Armando Falcucci, corresponding author)
The site
Grotta di Castelcivita and the Aurignacian stratigraphic sequence
Grotta di Castelcivita (hereafter Castelcivita; Salerno, Campania, Southern Italy) is a cave site situated at an elevation of 94 meters above sea level, nestled at the base of the Alburni massif, near the right bank of the Calore River (40.49563600N, 015.20922177E; see Fig. 1). This cave is part of a karst system that spans two primary levels, comprising an extensive network of tunnels and chambers, with a total length exceeding 5 kilometers (Cafaro et al., 2016). Systematic excavations were first conducted from 1975 to 1988, led by P. Gambassini of the Research Unit of Prehistory and Anthropology at the University of Siena (Gambassini, 1997). Subsequently, fieldwork was resumed in 2015 by the same research unit, under the direction of A. Ronchitelli and A. Moroni, and continuing to the present year, in collaboration with the “Soprintendenza Archeologia, Belle Arti e Paesaggio per le province di Salerno e Avellino”. The archaeological deposit is situated at the actual cave entrance and presently covers an excavated area of 35 square meters, of which Gambassini excavated 14 in the last century. Before the systematic investigations in 1975, nearly 6 square meters were excavated by looters (see Fig. 2b–d), leaving a deep large pit down to the base of the anthropogenic sequence. The destination of the archaeological materials originating from this area is presently unknown. The archaeological sequence spans a depth of 3.4 meters and contains multiple layers attributed to the late Mousterian (cgr, gar, lower-rsi), the Uluzzian (upper-rsi, pie, rpi, and rsa’‘), and the Aurignacian (rsa’, gic, and ars; see Fig. 2a). A noteworthy feature of this sedimentary sequence is the presence of a multi-layered flowstone with embedded thin layers of volcanic ashes that sealed the entire stratigraphic sequence. These layers of tephra have been identified as originating from the well-known Campanian Ignimbrite (CI) volcanic super-eruption (Fedele et al., 2004; Giaccio et al., 2008), and recently dated to 39.85 ± 0.14 ka BP (Giaccio et al., 2017). Consequently, the tephra represents a terminus ante quem for the site’s last human occupations. In this paper, our focus lies on the Aurignacian sequence, which follows the last Uluzzian layer rsa’’, dated to 41.9–40.6 ky cal BP (Douka et al., 2014; Wood et al., 2012). The Aurignacian cultural sequence is thus well-constrained chronologically between the last Uluzzian and the CI (39.85 ± 0.14 ka BP). This suggests that layers rsa’–ars accumulated over a relatively short period, likely just a few centuries (Giaccio et al., 2008). Castelcivita is, therefore, together with Grotta Paglicci in Gargano (Palma di Cesnola, 2004), a unique case study to investigate the initial stages and early development of the Aurignacian in southern Italy. The stratigraphic succession of Castelcivita was further revisited in 2020 by I. Martini, adopting a facies analysis-based approach. Layer rsa’ is made of fine-grained reddish sediments dominated by silt and sand with scattered small limestone debris (2-4 cm in size). The limestone debris results from the cave’s roof degradation, while the sandy sediments are linked to infiltration processes (sensu Martini et al., 2018; Martini et al., 2021) from the outer area. This layer is 12–15 cm thick and, generally, in direct continuity with the Uluzzian from layer rsa’’, except for a few discontinuous thin sterile sandy lenses, separating the two levels in some areas. The subsequent layer, gic, has a thickness of around 20 cm and is well distinguishable from rsa’ due to its yellowish color and the presence of extensive concretions. Finally, layer ars was excavated over a very limited area because the upper sequence was almost completely disrupted during the construction of an artificial entrance for visitor access to the cave. The sediment composition is similar to that of rsa’, primarily sandy but with a lighter color (i.e., orange). The excavations revealed a few features across the Aurignacian sequence. In the area excavated by Gambassini, a part of an extended surface with fire was identified at the base of layer gic, in square H14 between 80 and 90 cm deep. The remaining portion of this surface was brought to light in 2016 in square H15, thus attesting to the presence of a large or two smaller adjacent fireplaces (Supplementary Fig. S1). The fireplace/s, around 2x1 m in total, was/were composed of a layer of ash (above) and a layer of charcoal (below) and was/were lying directly on the surface of the ground, partially in contact with the underlying layer rsa’. Gambassini’s excavations did not report any traces of fireplaces in layer rsa’, except for a cluster of charcoal found in square H13. However, recent investigations have also revealed similar features to gic in this layer. The excavations conducted between 1975 and 1988 employed a stratigraphic method. Findings were meticulously documented using a grid coordinate system, which consisted of square meters further subdivided into 50x50 cm sectors. Layers were systematically excavated using 10 cm deep spits, further divided into 5 cm sub-spits. Excavators paid close attention to identified discontinuities within each spit, allowing them to follow sloping deposits and anthropogenic features accurately. All archaeological findings were precisely assigned to a specific square, sector, layer, and sub-spit. Furthermore, a significant portion of the materials were spatially documented in three dimensions using elevation (Z) coordinates. All sediments were carefully dried sieved and subsequently subjected to wet sieving with a 1 mm mesh. Further screening was then performed to isolate and categorize all finds.
Figure S1. Plan view of the excavation of layer
gic (spit 8) depicting evidence of the large anthropogenic
feature/s (in black) identified in squares H14 and H15. The other
excavated squares are colored yellow. Please, refer to the provided
legend for interpreting the drawing.
The environmental and ecological setting
The environmental and ecological setting of Castelcivita was comprehensively examined by studying large and small mammals, avifauna, ichtyofauna, anthracological remains, and the sedimentary composition of the sequence (see papers in Gambassini, 1997). The evidence gathered highlights significant changes during the transition from rsa’ to gic. Layer rsa’ contains a fauna assemblage that is notably similar to the last Uluzzian of layer rsa’’, marked by the predominance of horses and a significant presence of Microtus arvalis/agrestis and Microtus (Terricola) savii. Even if rare, taxa related to milder or more humid conditions are still present (e.g. fallow deer, Apodemus sp. or Elyomys). The avifauna, on the other hand, is characterized by a high frequency of steppe grassland species, particularly those associated with rocky environments. These findings, alongside the sedimentary composition of the layer, collectively suggest relatively cold and arid conditions, featuring open environments with sparse woodlands. Climatic features do not look to be extreme, as indicated by the presence of taxa more related to Mediterranean conditions. In layer gic, there is a discernible shift in climate towards a more humid, cold-temperate environment. This is accompanied by a reduction in the presence of the horse and an increase in red deer, roe deer, and chamois. Additionally, there is a slight rise in forest and water bird species (Fiore et al., 2020; Gambassini, 1997). Finally, layer ars records a new cold phase, evident particularly in the composition of anthracological remains dominated by Pinus and Betula. Integrated studies are ongoing to establish a correlation between the local environmental signal detected across the stratigraphic sequence and supra-regional climatic changes. Some evidence seems to indicate that Heinrich Event 4 (H4) commenced slightly before the deposition of the CI tephra (Badino et al., 2020; Lowe et al., 2012; Margari et al., 2009; Wulf et al., 2018; Wutke et al., 2015). In northern Italy, such as at Fumane Cave, H4 has been identified based on micro- and macro-fauna evidence (López-García et al., 2015). Marín-Arroyo et al. (2023), for instance, associated the reduced frequency of red deer in layer D3 with the onset of H4. Although the fauna assemblage composition may reflect hunting choices made by Aurignacian foragers (Discamps et al., 2011), significant oscillations in the frequency of red deer and horse at Castelcivita likely indicate changes in the surrounding environment of the cave. It is noteworthy, however, that red deer exhibit great ecological plasticity and can adapt to steppe environments, though they prefer patchy wooded areas (Discamps et al., 2011). At Castelcivita, two cold phases could be associated with the onset of H4. The first is linked with the driest phase of the sequence detected in the late Uluzzian (rsa’’) and the Protoaurignacian (rsa’) layers. Despite internal climatic oscillations being recorded across the period of H4 (Margari et al., 2009; Skinner & Elderfield, 2007), the amelioration detected in layer gic is rather marked and suggests that rsa’’–rsa’ accumulated more likely during the short-lived cold stadial GS 9/10, with layer gic corresponding to the GI 9 (Andersen et al., 2006; Rasmussen et al., 2014; Svensson et al., 2008). In this framework, H4 would only start in the uppermost layer ars. While an accurate taphonomic study is still in progress, we can state that humans played a significant role in the accumulation of mammal bones in the Aurignacian layers. Notably, evidence of carnivore activity, such as gnawing marks, is relatively low in comparison to the preceding Mousterian and Uluzzian (Romandini et al., 2020). This pattern slightly differs for birds, where there are fewer anthropogenic modifications, but several marks produced by carnivores. It is important to consider that the reduced exploitation of birds and fishes compared to the Uluzzian layers may be linked to differences in site use, a factor that warrants further exploration (Fiore et al., 2020).
Results
Raw material variability
The most prevalent raw material at Castelcivita is the local fine-grained chert, with comparable frequencies (approximately 90%) observed across the sequence. In our studied sample, quartzite, radiolarite, and coarse-grained chert were utilized at relatively low frequencies (Tables S1 and S2). Raw materials could be procured in the form of large-sized blocks from primary or sub-primary sources in the vicinity of the cave, as well as river pebbles, which were locally available along the riverbed of the Calore stream (Gambassini, 1997; Rossini et al., 2022). Unworked raw material blocks were not uncovered within the excavated area. Both tested (n = 2) and initial (n = 17) cores suggest that foraging groups opted to transport thick flakes and chunks, likely having shattered the raw material blocks away from the site.
| Layer | |||
|---|---|---|---|
| Raw material | ars | gic | rsa’ |
| Coarse-grained chert | 3 (3.0%) | 17 (1.4%) | 34 (2.6%) |
| Fine-grained chert | 89 (88.1%) | 1,125 (92.5%) | 1,193 (91.5%) |
| Limestone | 1 (1.0%) | 0 (0.0%) | 0 (0.0%) |
| Quartzite | 6 (5.9%) | 32 (2.6%) | 59 (4.5%) |
| Radiolarite | 2 (2.0%) | 42 (3.5%) | 18 (1.4%) |
| Total | 101 (100.0%) | 1,216 (100.0%) | 1,304 (100.0%) |
Table S1. Distribution of blanks and tools categories sorted according to raw material type. Rounded percentages are given in brackets.
| Layer | |||
|---|---|---|---|
| Raw material | ars | gic | rsa’ |
| Coarse-grained chert | 0 (0.0%) | 1 (1.1%) | 10 (8.1%) |
| Fine-grained chert | 8 (88.9%) | 89 (94.7%) | 105 (85.4%) |
| Limestone | 1 (11.1%) | 1 (1.1%) | 2 (1.6%) |
| Quartzite | 0 (0.0%) | 1 (1.1%) | 5 (4.1%) |
| Radiolarite | 0 (0.0%) | 2 (2.1%) | 1 (0.8%) |
| Total | 9 (100.0%) | 94 (100.0%) | 123 (100.0%) |
Table S2. Distribution of core and core-tool categories sorted according to raw material type. Rounded percentages are given in brackets.
Flake production and bipolar technology
Flakes were produced in all analyzed assemblages using platform, multidirectional, and bipolar strategies (Table S3). The latter are particularly attested in rsa’ and sharply decrease in the following layer gic (Fig. S2).
| Layer | Bipolar | Platform | Multidirectional | Shatter | Total |
|---|---|---|---|---|---|
| ars | 1 (50.0%) | 1 (50.0%) | 0 (0.0%) | 0 (0.0%) | 2 (100.0%) |
| gic | 15 (41.7%) | 2 (5.6%) | 12 (33.3%) | 7 (19.4%) | 36 (100.0%) |
| rsa’ | 48 (71.6%) | 9 (13.4%) | 7 (10.4%) | 3 (4.5%) | 67 (100.0%) |
| Total | 64 (61.0%) | 12 (11.4%) | 19 (18.1%) | 10 (9.5%) | 105 (100.0%) |
Table S3. Core types associated with the production of flakes across the studied assemblages. The category Shatter contains fragments that do not retain any evidence for laminar productions. The table does not include tested cores (n = 2) as the production objective could be not assessed. Rounded percentages are given in brackets.
Figure S2. Bipolar cores found in layers gic
(a) and rsa’ (b-f). The numbers shown after the letters are
from the dataset created by one of us (AF). In both the dataset and the
3D model repository, they are prefixed with “CTC”, which is the common
abbreviation for the site.
Most of the bipolar cores have two opposed striking platforms (Table S4) and in a few cases cores were rotated to carry on the blank production from orthogonal platforms. One of the most recurrent features of bipolar cores is the presence of at least a dihedral platform, which likely results from the disintegration of the core’s platform after several bipolar strikes (Arrighi et al., 2020; Peresani et al., 2019). Given that this platform type is usually found in only one side of the core, it has been supposed that this area was in contact with the anvil throughout the reduction (Arrighi et al., 2020). Scars of negatives can be typed in most cases as flakes and many show stepped or hinged terminations. Only in a limited number of cases bladelet-like scars are visible alongside with flake removals (n = 9, 7 in rsa’ and 2 in gic). Overall, there is no clear intention to produce bladelets in this class of cores and evidence of bipolar knapping is extremely low among the analyzed laminar blanks. Likewise, tools made on bipolar blanks are extremely rare (rsa’ = 6, gic = 1, and ars = 1) and never associated with tool types such as endscrapers, burins, and retouched blades or bladelets .
| Number of platforms | |||||
|---|---|---|---|---|---|
| Layer | 1 | 2 | 3 | 4 | Total |
| ars | 0 (0.0%) | 1 (100.0%) | 0 (0.0%) | 0 (0.0%) | 1 (100.0%) |
| gic | 1 (6.7%) | 13 (86.7%) | 0 (0.0%) | 1 (6.7%) | 15 (100.0%) |
| rsa’ | 5 (10.4%) | 39 (81.2%) | 4 (8.3%) | 0 (0.0%) | 48 (100.0%) |
Table S4. Number of striking platforms recorded on bipolar cores in the studied assemblages. Rounded percentages are given in brackets.
A comparison of the 3D volume of freehand and bipolar cores in both rsa’ and gic shows that bipolar cores have significantly lower values (Fig. S3), suggesting that this reduction technique allowed knappers to maximize blank production and exhaust most of the volume available. At the same time, the use of this technique can result in the split of the core in two or more bipolar shatters that are likely to be classified as bipolar cores due to the difficulty in finding a clear separation between them. This would also explain the high frequency of bipolar cores compared to freehand flake cores. Likewise, bipolar technique could also be used in an advanced stage of reduction to maximize blank production. In this regard, bipolar cores preserve less often cortical remains compared to platform and multidirectional flake cores (Table S5).
Figure S3. Comparison of the volume values of freehand
(i.e., platform and multidirectional) and bipolar cores in gic
and rsa’. Layer ars is not displayed as only one
bipolar core is available. The figure displays also the results of the
Wilcoxon tests comparing the volume values of freehand and bipolar cores
within each layer.
| Core type | ||||
|---|---|---|---|---|
| Cortex | Bipolar | Platform | Multidirectional | Shatter |
| 0% | 48 (75.0%) | 2 (16.7%) | 6 (31.6%) | 4 (40.0%) |
| 1-33% | 10 (15.6%) | 6 (50.0%) | 7 (36.8%) | 2 (20.0%) |
| 33-66% | 2 (3.1%) | 4 (33.3%) | 5 (26.3%) | 4 (40.0%) |
| 66-99% | 4 (6.2%) | 0 (0.0%) | 1 (5.3%) | 0 (0.0%) |
| Total | 64 (100.0%) | 12 (100.0%) | 19 (100.0%) | 10 (100.0%) |
Table S5. Percentage of cortex coverage recorded on flake cores considering all layers as a single group. Rounded percentages are given in brackets.
Initialization and configuration of blade and bladelet cores
Despite the advanced stage of reduction resulting in the discard of most cores, we managed to identify the blank types selected for knapping laminar blanks. Knappers typically selected block chunks, pebbles, and thick flakes for laminar production (Table S6). Striking platforms are consistently plain and were created by either using core tablets or positioning them on a ventral face when a flake was chosen. Faceted platforms are on the other hand absent. The main operations observed on initial cores primarily involved the decortication and shaping of the longitudinal and transversal convexities, typically executed through unidirectional strategies. Blank production often began by removing fully cortical or dihedral blanks, making use of sharp natural angles. Primary crests are also documented, and in most cases, they are one-sided, indicating that only one flank of the core was shaped with orthogonal removals. Crested blanks are relatively rare in gic (n = 3) and ars (n = 1), while they are more common in rsa’ (n = 25).
| Selected blank | ||||||||
|---|---|---|---|---|---|---|---|---|
| Layer | Angular debris | Blade | Block | Core fragment | Flake | Pebble | Undetermined | Total |
| ars | 0 (0.0%) | 0 (0.0%) | 1 (14.3%) | 0 (0.0%) | 4 (57.1%) | 1 (14.3%) | 1 (14.3%) | 7 (100.0%) |
| gic | 3 (5.2%) | 1 (1.7%) | 4 (6.9%) | 1 (1.7%) | 36 (62.1%) | 4 (6.9%) | 9 (15.5%) | 58 (100.0%) |
| rsa’ | 0 (0.0%) | 2 (3.7%) | 17 (31.5%) | 3 (5.6%) | 14 (25.9%) | 6 (11.1%) | 12 (22.2%) | 54 (100.0%) |
Table S6. Classification of cores according to the identified blank used for blade and bladelet productions. The category Undetermined includes all cores that do not retain enough information to identify the blank selected. Rounded percentages are given in brackets.
Production and maintenance operations on blade and bladelet cores
Carinated cores
Maintenance operations on carinated cores typically resulted in wide and convex flakes, often exhibiting bladelet negatives on the dorsal side. These operations aimed to isolate the flaking surface and maintain its transversal convexities, while also removing areas of the flaking surface with hinged removals. Such blanks have been identified in assemblages characterized by the presence of carinated technology (Kolobova et al., 2014; Le Brun-Ricalens, 2005). In total, we identified 110 blanks used for maintaining carinated cores. Notably, the majority of these blanks are from layer gic (n = 78, 71%). We compared the lengths of these blanks with the flaking surface of carinated cores in both gic and rsa’, observing a general intra-layer agreement (Fig. S5). This finding supports the specificity and stratigraphic attribution of carinated technology to all studied assemblages.
Figure S4. Comparison of the length of flaking surfaces
of carinated cores across the studied sequence. The figure displays the
results of the Kruskal-Wallis test and the pairwise comparisons.
Figure S5. Boxplots showing the distribution of length
values of the flaking surfaces of carinated cores and the length of
blanks identified as belonging to the maintenance of carinated cores.
The figure also displays the results of the Wilcoxon tests comparing
these values within gic and rsa’.
Platform cores
Bladelets are the predominant production goal across the sequence. Notably, independent blade production is only observed in gic, while simultaneous blade-bladelet production is most evident in rsa’. Simultaneous productions can also be inferred from the presence of blades with visible bladelet scars on their dorsal sides (Bon & Bodu, 2002). These scars indicate either the detachment of bladelets during blade reduction sequences (i.e., when the core allowed for blade production) or the removal of large blanks from bladelet cores, primarily for maintaining their convexities (Falcucci et al., 2017). At Castelcivita, 38 out of 42 blades with bladelet negatives identified relate to maintenance operations on bladelet cores, with the majority from layer rsa’ (71%, n = 30). This combined evidence is not surprising when considering that layer gic is primarily defined by the use of carinated technology, which rarely results in the detachment of maintenance blades (Le Brun-Ricalens, 2005).
PCA of blade and bladelet cores
Figure S6. Visualization of the results of the first
and third components of the PCA conducted on laminar cores.
A shows a biplot with the contribution of the different
quantitative variables to the first and second components.
B and C display the distribution of
the studied cores in the PC1 to PC3 space, sorted according to layer
(B) and core classification (C). In
A, FSL stands for flaking surface length,
FSL/T is the ratio between flaking surface length and
thickness, FSL/W is the ratio between flaking surface length
and width. The category Narrow/Burin includes narrow-sided cores and
burin cores. Initial cores were excluded from the analysis.
Morphometric analysis of blades and bladelets
In this section, we will explore the morphometric analysis of complete blades and bladelets to further delve into technological variability across the sequence.
Blades
The number of available blades is relatively low compared to bladelets, primarily due to blade production not being a primary goal at the site. Constraints imposed by locally available raw materials may have played a role in that. Nonetheless, we have a statistically suitable sample for conducting a morphometric analysis of layers gic and rsa’. In both Layers, blades were produced using direct marginal percussion (Tables S7-S10 and Fig. S7). The platforms are generally plain, and their comparable dimensions suggest a uniform knapping technique. The presence of lipped internal platform edges and moderately marked bulbs suggests the use of soft hammers, whether mineral or organic. Blade production is characterized by predominantly unidirectional sub-parallel removals, with bidirectional scars being rare. Differences were not found in profile curvature, profile twisting, and blank shape (Tables S11-S14). On the other hand, the study of cross-sections suggests that the increased frequency of trapezoidal and, to a lesser extent, polyhedral shapes are in part to be linked to the frequent use of blades in rsa’ for maintaining bladelet cores (Table S15). Finally, elongation (length to width ratio) and robustness (width to thickness ratio) ratios remain stable across gic and rsa’ (Table S16, Figs. S8 and S9), whereas linear measurements show that blades from gic are shorter and narrower (Table S17 and Fig S10).
Tables and figures reporting attributes linked to the knapping technique
| Layer | ||
|---|---|---|
| Platform type | gic | rsa’ |
| Plain | 33 (76.7%) | 44 (67.7%) |
| Linear/Punctiform | 4 (9.3%) | 7 (10.8%) |
| Other | 5 (11.6%) | 10 (15.4%) |
| Undetermined | 1 (2.3%) | 4 (6.2%) |
| Total | 43 (100.0%) | 65 (100.0%) |
Table S7. Platform types recorded on blades from rsa’ and gic. The category Other includes categories found in low frequencies (e.g., cortical, dihedral, double, abraded). Linear and punctiform types are grouped in a single category. A Fisher’s Exact Test reveals no differences between layers (p = 0.78).
| Layer | variable | n | mean | sd | min | median | max |
|---|---|---|---|---|---|---|---|
| gic | Platform_width | 41 | 4.188 | 2.683 | 0.2 | 3.80 | 12.5 |
| gic | Platform_thickness | 41 | 2.037 | 1.576 | 0.2 | 1.70 | 6.4 |
| rsa’ | Platform_width | 60 | 4.760 | 3.517 | 0.2 | 3.35 | 15.8 |
| rsa’ | Platform_thickness | 60 | 2.188 | 1.942 | 0.1 | 1.50 | 8.1 |
Table S8. Summary statistics (in mm) of the width and thickness measurements recorded on blades. SD stands for standard deviation.
Figure S7. Boxplots showing the distribution of
platform width (A) and thickness (B)
values in gic and rsa’. The figure is complemented by
the results of the performed Wilcoxon tests, confirming the marked
similarity of these attributes between layers.
| Lip type | ||||
|---|---|---|---|---|
| Layer | Absent | Moderate | Pronounced | Total |
| gic | 19 (44.2%) | 9 (20.9%) | 15 (34.9%) | 43 (100.0%) |
| rsa’ | 32 (49.2%) | 13 (20.0%) | 20 (30.8%) | 65 (100.0%) |
Table S9. Presence and type of lips recorded on blades. A Fisher’s Exact Test reveals no differences between layers (p = 0.87).
| Bulb type | ||||
|---|---|---|---|---|
| Layer | Absent | Moderate | Pronounced | Total |
| gic | 15 (34.9%) | 23 (53.5%) | 5 (11.6%) | 43 (100.0%) |
| rsa’ | 22 (33.8%) | 36 (55.4%) | 7 (10.8%) | 65 (100.0%) |
| Total | 37 (34.3%) | 59 (54.6%) | 12 (11.1%) | 108 (100.0%) |
Table S10. Presence and type of bulbs recorded on blades. Fisher’s Exact Test reveals no differences between layers (p = 1).
Tables and figures reporting technological and morphological attributes
| Scar pattern | |||||
|---|---|---|---|---|---|
| Layer | Unidirectional parallel | Unidirectional convergent | Bidirectional | Other | Total |
| gic | 24 (55.8%) | 13 (30.2%) | 3 (7.0%) | 3 (7.0%) | 43 (100.0%) |
| rsa’ | 28 (43.1%) | 19 (29.2%) | 6 (9.2%) | 12 (18.5%) | 65 (100.0%) |
Table S11. Scar patterns recorded on the blade assemblages. The Other category includes scar patterns found in low frequencies (e.g., crossed, unidirectional transverse, and undetermined patterns). The result of a Fisher’s Exact Test reveals no differences between layers (p = 0.34).
| Curvature | ||||
|---|---|---|---|---|
| Layer | Curved | Curved slightly | Straight | Total |
| gic | 22 (51.2%) | 6 (14.0%) | 15 (34.9%) | 43 (100.0%) |
| rsa’ | 32 (49.2%) | 18 (27.7%) | 15 (23.1%) | 65 (100.0%) |
Table S12. Presence and intensity of profile curvature recorded on complete blades. The result of a Fisher’s Exact Test reveals no differences between layers (p = 0.16).
| Torsion simplified | |||
|---|---|---|---|
| Layer | no | yes | Total |
| gic | 29 (67.4%) | 14 (32.6%) | 43 (100.0%) |
| rsa’ | 42 (64.6%) | 23 (35.4%) | 65 (100.0%) |
Table S13. Presence of profile twisting recorded on complete blades. The result of a Fisher’s Exact Test reveals no differences between layers (p = 0.84).
| Blank shape | |||||
|---|---|---|---|---|---|
| Layer | Sub-parallel | Convergent | Irregular | Other | Total |
| gic | 19 (55.9%) | 2 (5.9%) | 9 (26.5%) | 4 (11.8%) | 34 (100.0%) |
| rsa’ | 22 (37.3%) | 9 (15.3%) | 22 (37.3%) | 6 (10.2%) | 59 (100.0%) |
Table S14. External shape recorded on blades. The category Other includes categories found in low frequencies (e.g., convex, comma-like). A Fisher’s Exact Test reveals no differences between layers (p = 0.27).
| Cross-section | |||||
|---|---|---|---|---|---|
| Layer | Lateral steep | Polyhedral | Trapezoidal | Triangular | Total |
| gic | 11 (25.6%) | 9 (20.9%) | 11 (25.6%) | 12 (27.9%) | 43 (100.0%) |
| rsa’ | 12 (18.5%) | 20 (30.8%) | 30 (46.2%) | 3 (4.6%) | 65 (100.0%) |
Table S15. Cross-section shape recorded on blades. A Fisher’s Exact Test reveals significant differences between layers (p = 0.002).
| Layer | variable | n | mean | sd | min | median | max |
|---|---|---|---|---|---|---|---|
| gic | Elongation | 34 | 2.5 | 0.4 | 2.0 | 2.4 | 3.8 |
| gic | Robustness | 34 | 3.2 | 1.0 | 1.6 | 3.1 | 5.9 |
| rsa’ | Elongation | 59 | 2.6 | 0.5 | 2.0 | 2.5 | 4.0 |
| rsa’ | Robustness | 59 | 3.3 | 1.3 | 1.5 | 3.0 | 7.2 |
Table S16. Summary statistics of the elongation (length to width ratio) and robustness (width to thickness ratio) of blades. SD stands for standard deviation.
Figure S8. Boxplots showing the distribution of
elongation (A) and robustness (B)
ratios in gic and rsa’. The figure is complemented by
the results of the performed Wilcoxon tests, confirming the similarity
of these attributes between layers, especially in relation to
robustness.
Tables and figures reporting metric attributes
| Layer | variable | n | mean | sd | min | median | max |
|---|---|---|---|---|---|---|---|
| gic | Length | 34 | 36.0 | 7.9 | 25.0 | 36.0 | 59.6 |
| gic | Width | 34 | 14.6 | 2.4 | 12.1 | 13.6 | 20.8 |
| gic | Thickness | 34 | 5.0 | 1.9 | 2.8 | 4.6 | 10.5 |
| rsa’ | Length | 59 | 42.5 | 10.3 | 24.7 | 42.7 | 65.5 |
| rsa’ | Width | 59 | 16.4 | 4.0 | 12.3 | 15.7 | 31.6 |
| rsa’ | Thickness | 59 | 5.8 | 2.9 | 2.0 | 4.9 | 16.2 |
Table S17. Summary statistics of linear dimensions (length, width, and thickness in mm) recorded on complete blades, excluding those modified by lateral retouch. SD stands for standard deviation.
Figure S9. Boxplots showing the distribution of length
(A), width (B), and thickness
(C) in gic and rsa’. The figure is
complemented by the results of the performed Wilcoxon tests, showing
that blades from gic are shorter and narrower.
Bladelets
In contrast to blades, bladelets exhibit more noticeable variations across the sequence, and the larger number of artifacts found in ars allows us to include this layer in the comparison. Bladelet blanks were still detached using direct freehand knapping, but in gic and ars, the motion appears to have been more marginal compared to rsa’, as visible from the increased presence of linear and punctiform platforms (Tables S18-S21). Notably, significant differences in platform width and thickness are observed between rsa’ and the upper layers (Fig. S10). Bulbs are more frequently absent in gic and especially ars, whereas lipped internal platforms are more frequent. Further experimental work is required to determine if these differences are related to distinct knapping techniques used in carinated core reduction. In terms of the flaking direction recorded on the visible scars of bladelets, it is almost always unidirectional (Table S22). In gic, reduction pattern is more frequently convergent, although no differences were identified in the external morphology and distal ends in dorsal view (Tables S23 and S24, but see 2DGM analysis). Profiles are straighter in gic, while profile twisting is more common in rsa’ (Tables S25-S26). In layers ars and gic, bladelet cross-sections are often triangular (Table S27), suggesting a preference for a single core ridge to guide removal (see below). The elongation ratio confirms significant morphological variability between the upper layers and rsa’, whereas the robustness remains relatively consistent throughout the sequence (Table S28 Fig. S11). A more detailed exploration of these morphological aspects through a shape analysis, including retouched and unretouched specimens, is presented after the tool analysis. As depicted in Fig. S12, the bladelets recovered in gic and ars are smaller in terms of length, width, and thickness compared to those from rsa’ (Table S29). The differences in length values are particularly pronounced and can be associated with the increased use of carinated technology.
Tables and figures reporting attributes linked to the knapping technique
| Layer | |||
|---|---|---|---|
| Platform type | ars | gic | rsa’ |
| Plain | 15 (34.9%) | 199 (35.3%) | 140 (50.5%) |
| Linear | 18 (41.9%) | 252 (44.7%) | 87 (31.4%) |
| Punctiform | 9 (20.9%) | 107 (19.0%) | 32 (11.6%) |
| Other | 1 (2.3%) | 3 (0.5%) | 15 (5.4%) |
| Undetermined | 0 (0.0%) | 3 (0.5%) | 3 (1.1%) |
| Total | 43 (100.0%) | 564 (100.0%) | 277 (100.0%) |
Table S18. Platform types recorded on bladelets. The category Other includes categories found in low frequencies (e.g., cortical, dihedral, double, abraded). A Pearson’s chi-squared test reveals significant differences between layers (Chi-squared = Chi=47.56, p < 0.01).
| Layer | variable | n | mean | sd | min | median | max |
|---|---|---|---|---|---|---|---|
| ars | Platform_width | 43 | 1.581 | 1.186 | 0.1 | 1.5 | 5.7 |
| ars | Platform_thickness | 43 | 0.567 | 0.616 | 0.1 | 0.2 | 2.7 |
| gic | Platform_width | 561 | 1.580 | 1.027 | 0.1 | 1.6 | 7.5 |
| gic | Platform_thickness | 561 | 0.484 | 0.528 | 0.1 | 0.2 | 4.9 |
| rsa’ | Platform_width | 271 | 2.060 | 1.358 | 0.1 | 2.0 | 10.0 |
| rsa’ | Platform_thickness | 271 | 0.788 | 0.794 | 0.1 | 0.7 | 6.0 |
Table S19. Summary statistics (in mm) of the width and thickness measurements recorded on bladelets. SD stands for standard deviation.
Figure S10. Boxplots showing the distribution of
platform width (A) and thickness (B)
values across the studied sequence. The figure includes results of the
Kruskal-Wallis test and the pairwise comparisons. Statistically
significant differences are observed when comparing both ars
and gic to rsa’.
| Bulb type | ||||
|---|---|---|---|---|
| Layer | Absent | Moderate | Pronounced | Total |
| ars | 35 (81.4%) | 7 (16.3%) | 1 (2.3%) | 43 (100.0%) |
| gic | 355 (62.9%) | 200 (35.5%) | 9 (1.6%) | 564 (100.0%) |
| rsa’ | 149 (53.8%) | 123 (44.4%) | 5 (1.8%) | 277 (100.0%) |
| Total | 539 (61.0%) | 330 (37.3%) | 15 (1.7%) | 884 (100.0%) |
Table S20. Presence and type of bulbs recorded on bladelets. A Pearson’s chi-squared test reveals significant differences between layers (Chi-squared = 15.14, p = 0.004).
| Lip type | ||||
|---|---|---|---|---|
| Layer | Absent | Moderate | Pronounced | Total |
| ars | 2 (4.7%) | 32 (74.4%) | 9 (20.9%) | 43 (100.0%) |
| gic | 63 (11.2%) | 419 (74.3%) | 82 (14.5%) | 564 (100.0%) |
| rsa’ | 48 (17.3%) | 183 (66.1%) | 46 (16.6%) | 277 (100.0%) |
Table S21. Presence and type of lips recorded on bladelets. A Pearson’s chi-squared test reveals significant differences between layers (Chi-squared = 11.02, p = 0.03).
Tables and figures reporting technological and morphological attributes
| Scar pattern | |||||
|---|---|---|---|---|---|
| Layer | Unidirectional parallel | Unidirectional convergent | Bidirectional | Other | Total |
| ars | 23 (53.5%) | 18 (41.9%) | 0 (0.0%) | 2 (4.7%) | 43 (100.0%) |
| gic | 237 (42.0%) | 308 (54.6%) | 5 (0.9%) | 14 (2.5%) | 564 (100.0%) |
| rsa’ | 138 (49.8%) | 119 (43.0%) | 4 (1.4%) | 16 (5.8%) | 277 (100.0%) |
Table S22. Scar patterns recorded on the bladelet assemblages. The Other category includes scar patterns found in low frequencies (e.g., crossed, unidirectional transverse, and undetermined patterns). The result of a Pearson’s chi-squared test reveals significant differences between layers (Chi-squared = 15.61, p = 0.02).
| Blank shape | |||||
|---|---|---|---|---|---|
| Layer | Sub-parallel | Convergent | Irregular | Other | Total |
| ars | 15 (39.5%) | 12 (31.6%) | 3 (7.9%) | 8 (21.1%) | 38 (100.0%) |
| gic | 185 (40.8%) | 160 (35.3%) | 53 (11.7%) | 55 (12.1%) | 453 (100.0%) |
| rsa’ | 117 (46.4%) | 81 (32.1%) | 37 (14.7%) | 17 (6.7%) | 252 (100.0%) |
Table S23. External shape recorded on blades. The category Other includes categories found in low frequencies (e.g., convex, comma-like). A Pearson’s chi-squared test reveals no differences between layers (Chi-squared = 11.98, p = 0.06).
| Distal end | |||||
|---|---|---|---|---|---|
| Layer | Convex | Irregular | Pointed | Straight | Total |
| ars | 15 (39.5%) | 1 (2.6%) | 14 (36.8%) | 8 (21.1%) | 38 (100.0%) |
| gic | 163 (36.0%) | 39 (8.6%) | 181 (40.0%) | 70 (15.5%) | 453 (100.0%) |
| rsa’ | 91 (36.1%) | 16 (6.3%) | 94 (37.3%) | 51 (20.2%) | 252 (100.0%) |
Table S24. Distal end shape (in dorsal view) recorded on bladelet. A Pearson’s chi-squared test reveals no differences between layers (Chi-squared = 5.28, p = 0.51).
| Curvature | ||||
|---|---|---|---|---|
| Layer | Curved | Curved slightly | Straight | Total |
| ars | 18 (41.9%) | 11 (25.6%) | 14 (32.6%) | 43 (100.0%) |
| gic | 166 (29.4%) | 166 (29.4%) | 232 (41.1%) | 564 (100.0%) |
| rsa’ | 111 (40.1%) | 74 (26.7%) | 92 (33.2%) | 277 (100.0%) |
Table S25. Presence and intensity of profile curvature recorded on complete bladelet. The result of a Pearson’s chi-squared test reveals significant differences between layers (Chi-squared = 11.32, p = 0.02).
| Torsion simplified | |||
|---|---|---|---|
| Layer | no | yes | Total |
| ars | 39 (90.7%) | 4 (9.3%) | 43 (100.0%) |
| gic | 453 (80.3%) | 111 (19.7%) | 564 (100.0%) |
| rsa’ | 198 (71.5%) | 79 (28.5%) | 277 (100.0%) |
Table S26. Presence of profile twisting recorded on complete bladelets. The result of a Pearson’s chi-squared test reveals significant differences between layers (Chi-squared = 12.69, p = 0.002).
| Cross-section | ||||||
|---|---|---|---|---|---|---|
| Layer | Flat | Lateral steep | Polyhedral | Trapezoidal | Triangular | Total |
| ars | 4 (9.3%) | 5 (11.6%) | 1 (2.3%) | 18 (41.9%) | 15 (34.9%) | 43 (100.0%) |
| gic | 22 (3.9%) | 31 (5.5%) | 28 (5.0%) | 234 (41.5%) | 249 (44.1%) | 564 (100.0%) |
| rsa’ | 15 (5.4%) | 37 (13.4%) | 23 (8.3%) | 129 (46.6%) | 73 (26.4%) | 277 (100.0%) |
Table S27. Cross-section shape recorded on bladelets. A Pearson’s chi-squared test reveals significant differences between layers (Chi-squared = 38.74, p < 0.01).
| Layer | variable | n | mean | sd | min | median | max |
|---|---|---|---|---|---|---|---|
| ars | Elongation | 38 | 2.2 | 0.7 | 1.2 | 2.1 | 4.4 |
| ars | Robustness | 38 | 4.0 | 1.4 | 1.3 | 4.0 | 7.7 |
| gic | Elongation | 453 | 2.4 | 0.8 | 1.1 | 2.2 | 8.7 |
| gic | Robustness | 453 | 3.7 | 1.1 | 1.0 | 3.7 | 8.3 |
| rsa’ | Elongation | 252 | 2.9 | 0.9 | 1.1 | 2.7 | 6.9 |
| rsa’ | Robustness | 252 | 3.5 | 1.3 | 0.8 | 3.4 | 7.4 |
Table S28. Summary statistics of the elongation (length to width ratio) and robustness (width to thickness ratio) of bladelets. SD stands for standard deviation.
Figure S11. Boxplots showing the distribution of
elongation (A) and robustness (B)
ratios across the studied sequence. The figure includes results of the
Kruskal-Wallis test and the pairwise comparisons. Statistically
significant differences are observed when comparing both ars
and gic to rsa’.
Tables and figures reporting metric attributes
| Layer | variable | n | mean | sd | min | median | max |
|---|---|---|---|---|---|---|---|
| ars | Length | 38 | 14.0 | 5.7 | 5.5 | 13.0 | 30.5 |
| ars | Width | 38 | 6.5 | 2.4 | 2.8 | 6.2 | 11.3 |
| ars | Thickness | 38 | 1.8 | 0.9 | 0.7 | 1.7 | 5.1 |
| gic | Length | 453 | 15.5 | 5.6 | 6.3 | 14.5 | 38.2 |
| gic | Width | 453 | 6.7 | 2.0 | 2.1 | 6.6 | 12.0 |
| gic | Thickness | 453 | 1.9 | 0.8 | 0.6 | 1.7 | 6.2 |
| rsa’ | Length | 252 | 20.9 | 7.7 | 6.9 | 19.4 | 44.2 |
| rsa’ | Width | 252 | 7.3 | 2.2 | 1.9 | 7.4 | 11.9 |
| rsa’ | Thickness | 252 | 2.4 | 1.4 | 0.7 | 2.1 | 9.1 |
Table S29. Summary statistics of linear dimensions (length, width, and thickness in mm) recorded on complete bladelets, excluding those modified by lateral retouch. SD stands for standard deviation.
Figure S12. Boxplots showing the distribution of length
(A), width (B), and thickness
(C) across the studied sequence. The figure includes
results of the Kruskal-Wallis test and the pairwise comparisons.
Statistically significant differences are observed when comparing both
ars and gic to rsa’.
Tools
Common tools
Figure S13. Selection of tools and core-tools from
rsa’. The number following the alphabetical list corresponds to
the ID assigned by AF during the techno-typological analysis (refer to
the provided dataset for details). The figure includes: multiple burin
on prepared platform (a), busked burin (b), carinated endscraper (c),
endscrapers (d-e), blade with Aurignacian retouch (f), retouched blade
(g), and retouched flake (h). Drawings are from Gambassini (1997).
